What is a "public safety" intervention?

Hello from Savannah, Georgia, where the Niskanen Center’s annual Criminal Justice Convening just wrapped. Many of my favorite people in one place — what a treat!

I’ve been thinking a lot recently about the value of casting a wide net when looking for scalable interventions to reduce crime. Yes, it is important to make police, courts, and prisons more effective. But if that’s all we’re doing, we’re missing a lot of opportunities — not least because preventing someone’s first crime is way easier than getting them off a criminal path once they’re on it.

If you look at programs in the education, health, housing, and job training spaces, you might notice that they’re targeting largely the same population the criminal justice space is serving: kids and adults struggling with some combination of poverty, family instability, trauma, mental illness, and addiction. Everyone is working to find scalable interventions to put these people on a better path and help them thrive. They might measure outcomes like high school graduation, college enrollment, employment, and wages. Those are important, but not the full story. If a program has put someone on a better path — to more education and higher earnings — it has likely helped them avoid a path to crime and incarceration. Those avoided costs should be considered too. Because incarceration is so expensive, any intervention that keeps someone out of jail or prison pays for itself very quickly. That is useful information for ROI-focused policymakers to know.

At Arnold Ventures (AV), I became a broken record when talking with my colleagues from other parts of the organization: please add crime outcomes to your housing/education/health projects!

In my last post, I began discussing ongoing, AV-funded projects that I’m excited about. Today, I’ll highlight a study of an intervention outside the traditional criminal justice space:

PIs: Janey Rountree, Till von Wachter, and Brian Blackwell

AV funding: $99,054, in Q1 of 2026.

Housing instability is a challenge for many people who are arrested and incarcerated. Could interventions that help people find safe, stable places to live prevent crime and criminal justice involvement? This study is one of several ongoing research projects testing this approach.

LA’s Homelessness Prevention Unit (HPU) was launched in 2021 to serve families at high risk of homelessness. In partnership with the California Policy Lab — a fantastic research organization based at UCLA and UC Berkeley — HPU staff use an algorithm to determine which families are at highest risk, then proactively offer them assistance. This includes financial assistance that scales with the household size — on average, about $6,500 to pay for rent, utilities, food, transportation, and similar costs. This is a large amount of money, but if it puts families on a better path it could achieve much bigger savings for the city down the road. (There is previous evidence that targeted cash transfers like this can indeed be cost-effective.) Because policymakers will want to compare these costs and benefits — and consider how the net benefit stacks up against other potential uses of those scarce public funds — a randomized controlled trial (RCT) was added in 2023 to quantify the program’s impact. High-risk families are randomly assigned to receive HPU assistance or not. And then, within the treated group, some families are randomly assigned to receive an extra $2,000. Random assignment of families across these three arms (control, baseline treatment, and the “extra cash” treatment) helps ensure that any differences in families’ future outcomes can be attributed to the HPU program, rather than some pre-existing differences across families.

Of particular interest to me, the PIs will measure this program’s effects on criminal justice involvement, including arrests and days incarcerated. This will allow them to measure the social cost savings of this program, in terms of avoided crime and reduced criminal justice spending, and compare those savings with the cost of the program.

Preliminary results are expected in 2027, with follow-up reports to be published in subsequent years.

In the latest episode of Fighting Crime, host Cristina Quinn digs into cutting-edge research on the impacts of police trainings. You might be surprised that we know very little about which police trainings are effective at improving police outcomes. Most studies that do exist simply survey cops on their way out of the training room, asking what they learned. What we really want to know is if they perform differently on the job as a result of the training. Researchers — in collaboration with forward-looking police departments — are just beginning to test trainings in this way. Check out “How to Train a Cop” to learn more:

And in Fighting Crime’s companion podcast, Field Notes, AV’s Julie James interviews Andy Potter from One Voice United (OVU). Andy spent his career as a corrections officer, and founded OVU to bring that field’s perspective to important policy conversations. In particular, he highlights the ongoing “corrections crisis,” where chronic understaffing in jails and prisons leads to terrible conditions for those living and working in those facilities. Andy makes a powerful case that if we care about public safety, we need to pay attention to this crucial — but often ignored — workforce. Check it out:

• I loved this episode of Clearer Thinking, with Nick Allardice from GiveDirectly. If you’re interested in impact-driven philanthropy — including how to think about maximizing ROI in a world that’s full of uncertainty — it’s worth a listen.

• Nick Thompson recently interviewed Eli Pariser on The Most Interesting Thing in AI. The whole episode is great, but Eli’s answer to the last question (at 39:45) has been on my mind for a few days now. His take: So much energy in the AI space goes into making super-moral AI agents that would never do bad things. But haven’t we learned from thousands of years of human history that that’s a fool’s errand? Perhaps it would be more productive to assume that there will be badly-behaved agents, and focus on building an infrastructure that deters and incapacitates them before they can do too much harm. Do we need a criminal justice system for AI agents? 🤯

• I’m enjoying Joanna Stern’s I Am Not a Robot. It’s a great source of inspiration as I finally have time to learn how to use AI myself. (Note: these newsletters are written entirely by me, for better or worse.)

I’m gradually putting my Probable Causation podcast archive on YouTube. If you missed any of the past 100+ episodes, this is your chance to catch up! Please like and share your favorites to help others find the show.

I’ll leave you with this throwback to my very first episode, with Chloe Gibbs in April 2019, about the intergenerational effects of Head Start programs on criminal behavior (another high-impact intervention outside the traditional criminal justice space!):